On 06/08/2017 10:56 AM, William A Rowe Jr wrote:
On Wed, Jun 7, 2017 at 9:44 PM, Dennis Clarke <dcla...@blastwave.org> wrote:
On 06/07/2017 08:16 PM, William A Rowe Jr wrote:
As I said... checking out and building 'your own tarball' on a
conventional system absolutely beats convincing AIX, Solaris
By "conventional" I think you mean "Linux" or anything with a reasonable
GNU toolchain and we all know that won't happen from IBM or Oracle. At
all. Lucky for me I have the latest perl, autoconf, m4 and such built
already and "python" still blows up neatly in any attempt to build from
sources. Don't even ask about PHP ver 7.x where code really gets
gnuified and ugly.
Correct, a typical dev environment with a rather complete toolchain.
Note: I am not being a barking dog. I really am concerned. Long term.
Can you define "toolchain"? What I mean to say is that there seems to
be a defacto standard that shifts around from year to year or at least
decade to decade. No real standard but just a "this is the stuff that
works today" with no real promise that it complies with any documented
spec anywhere. Once upon a time it would be COBOL and JCL scripts on ye
IBM mainframes followed by whatever people were using on CPM and then
DOS for a while. I think SCO UNIX and then BSD variations were all over
the place for a while followed by SunOS based on BSD and VAX VMS as well
as IBM AIX and HPUX from HP. This shifts around over the years. Still
no real long term standards but if it ain't GNUified then it won't work
is the general mode of operation these days. What is tomorrow? The
point I am making, if I am hand waving with a coffee in hand here, is
that I don't see any long term vision or do we just hope things work and
then there really is no spec nor standards? I am looking at a lot of
projects here and not just ASF based items.
I run into problems between software projects ALL the time. The latest
in life is that the recent releases of GNU Autoconf ( 2.69 today ) won't
pass a testsuite within itself if you also have recent Perl. Recent
Perl releases may be a case of "good luck" trying to get it to pass some
basic tests. It won't. At least not across a suite of systems and
architectures. I know that the world is shifting gears again and I have
seen this over the long haul over and over. The fad and the sporty way
to go today is GNUified everything and extensions flying all over C code
with new world "defacto" ideas like -std=gnu99 ( for some variation of
C99 with GNU extensions not in real C99 ) and we get a result that just
isn't portable nor long term stable. It may work today. For a very brief
window. Maybe five years. It won't work long term.[1]
I feel motivated to write this letter because I see a trend all over the
various open source projects where integration and long term stability
is difficult to see. Some are amazing such as the releases of Xterm as
well as the folks working on libgmp or ISC BIND. I know that I can take
the sources from ISC BIND out of the tarball box and throw it at the
most strict compilers with strict CFLAGS and it compiles. Clean. The
odd warning may be seen. The Apache HTTPD project and its related bits
are essential to the world we know today. The word really should be
critical. Perhaps there is some other fancy three dollar word but I can
not think of it at the moment. I suspect that the readers on this mail
list fully grok what I am saying. The most recent releases of APR and
APR-UTIL are the stepping stones needed for running a really awesome and
amazing web server. Getting that to build and be *trusted* to run
without core files flying or run away processes chewing vast hogs of
memory is a task best left for scientists and software engineers. I am
neither. However I do have a whole suite of systems across a stack of
architectures running a few different operating systems and I use some
compilers that are variations and versions of gcc as well as the Oracle
Studio suite which is really just ye old Sun Forte developer tools which
go back even further in years. At the very least I do have access to
commercial supported systems that are tightly compliant with some very
well documented specs. They are not a joy to work with. Not always. But
they sure are stable and seriously secure. Long term. The code that gets
released from the ASF is generally awesome in its portability. You can
toss it into an acid test and it stands up. That makes life really easy
to be a "user" that actually compiles from the sources.
My question here is, what is the long term vision? The corporate "cloud"
game players are all watching and getting ready to chew up as much of
the internet as possible. That transition is happening now. The end
point objective is that the user has no access to anything that "works"
and no access to "why" or "how". Just pay the bill and be happy. Open
source and "free" software isn't a long term vision in that future. So
what is the long term vision really? Because what works today won't
work tomorrow. For some definition of "tomorrow". A lot of eyes are
watching closely and they are going for the money which has nothing to
do with the user. My suggestion is make life easy for anyone that wants
to be a user. I throw donations at the FSF yearly and try to support a
pile of open source projects in some way and often it means I take the
code from the source tarball and push it across a stack of systems. Try
to file a bug report and try to follow up. At least try. So this most
recent release of APR and APR-UTIL is of great concern to me and I know
there are awesome people working on it. I just hope that when the actual
release does happen the ordinary user guy at the keyboard can compile it
from sources in a portable and long term stable way. How can I help?
Dennis
[1] recent Power arch systems are an issue running Debian/RHEL/AIX
